Audio Fingerprinting
Kevin Curran, Computer Lecturer - Magee College

Audio Fingerprinting

The Philips article referenced below gives the following scenario: Louise is driving her car and tuning the car radio for a channel of her liking. Suddenly she hears the best song she heard in months. She likes it so much that she would like to buy the song, but she doesn't know the song. Of course she has missed the announcement as well, so what are her options?

She can of course try to remember the name of the radio station as well as the time and then call the radio station at a later occasion. But of course, by the time Louise gets home, she has forgotten everything. Wouldn't it be easier if Louise, while driving her car, simply could take out her cell-phone, push a few buttons, hold the phone up for a few moments, waiting for message containing the title of the song she was listening to? And that at the same time the song title and additional information was also sent to her mailbox?'

The scenario above is now possible with a new technology called audio fingerprinting. The essence of audio fingerprinting is twofold. Firstly, it entails the ability to extract unique features from music. These features are loosely analogous to normal fingerprints for human beings and are from now on referred to as (audio) fingerprints. Secondly, it requires clever searching algorithms to compare these audio fingerprints to large databases of previously extracted audio fingerprints.

Links

Audio Fingerprinting article by Philips
  • J.A. Haitsma, Ton Kalker, J. Oostveen, , "Robust Audio Hashing for Content Identification, CBMI 2001, Second International Workshop on Content Based Multimedia and Indexing, September 19-21, 2001, Brescia, Italy.
cbmi01audiohashv1.0.pdf
  • T. Kalker, "Applications and Challenges for Audio Fingerprinting", presentation at the 111th AES Convention, NY, in the "Watermarking versus Fingerprinting" workshop, December 3, 2001.
audiofp_aes01_ppt.pdf

Task

ComParser is computer software that recognizes realtime audio fragments. It will be expanded with pseudo-score-following, musical analysis, improvisation algorithms, etc. The software is written in ANSI C as far as possible, sourcecode, documentation and binaries may be downloaded from http://www.hku.nl/~pieter/SOFT/CMP/, it is designed to run on all platforms.

Use ComParser to examine how feasible the follow scenario is:

1. Ad agency, X, needs to track and report where new advert for Ford Focus is broadcast on TV and Radio.
2. They sample the advert through the "audio fingerprint system", and then using the the broadcast monitoring platform from hookablemedia, they can track the advert using the fingerprint
3. A report is generated showing when the advert was played, on what station, between what programs (information taken from an EPG). (Further reports can be generated to compare advert from Toyota vs Ford, etc).

Any questions?

Home